Meta confirms Llama 3 is arriving within the next month

Llama 3 is Meta's upcoming next-generation open-source large language model, designed to serve as the foundation for generative AI assistants.

Its primary focus is addressing the major usability hurdle of Llama 2: excessive caution. While previous models frequently refused prompts, Llama 3 is tuned to answer a much wider range of questions, including controversial topics, with higher accuracy.

Meta Chief Product Officer Chris Cox noted that Llama 3 will eventually power multiple user-facing products across the company's ecosystem.

Expected specs and release timeline

Meta has not yet revealed exact parameter counts or shared functional demos for Llama 3.

However, reports indicate that the flagship configuration is expected to feature around 140 billion parameters. That is double the footprint of Llama 2's largest 70-billion-parameter release.

Speaking at an event in London on Tuesday, Nick Clegg, Meta's president of global affairs, stated that the initial rollout will begin within the next month, and potentially even sooner.

Following that first drop, Meta plans to release several distinct versions with varying capabilities and versatilities throughout the remainder of the year.

Other generative tools remain on hold for now. AI Research VP Joelle Pineau confirmed that Meta will not be releasing its Emu image generator yet. Chief AI scientist Yann LeCun also reiterated his skepticism about generative models, pointing to Joint Embedding Predictive Architecture (JEPA) as the real long-term direction.

Why it matters

For developers building self-hosted pipelines or fine-tuning open weights, Llama 3 is the most important release on the immediate horizon. If the 140B variant can match proprietary tiers while easing restrictions, it will reset the benchmark for open-source AI.

The clear bottleneck will be compute. Stepping up to 140B puts the unquantized weights far out of reach for consumer hardware. Most individual developers will want to wait for community quantizations or the smaller variants slated to follow in the rollout.